Abstract The loss function is a criterion to evaluate the learning quality of a deep convolutional neural network, which represents the gap between prediction and ground truth. However, as the most commonly used loss function in image classification tasks, Cross‐Entropy loss does not encourage the model to distinguish the similarity between features. In this work, the authors investigate inter‐class separability of similar features learnt by convolutional networks and propose a loss function called Error Refactor Loss (ER‐Loss). ER‐Loss is based on the error caused by convolutional networks; it can improve the inter‐class separability and is simple to implement and can easily replace the Cross‐Entropy loss. Compared with softmax loss, ER‐Loss adds a dynamic penalty item which can help ER‐Loss monitor the actual situation of model training and adjust the value of the penalty item according to model training. The ER‐Loss on CIFAR100 and part of ImageNet ILSVRC 2012 is evaluated and the experimental result showed that the ER‐Loss can improve the accuracy of the model.

Facial video big sensor data (BSD) is the core data of wireless sensor network industry application and technology research. It plays an important role in many industries, such as urban safety management, unmanned driving, senseless attendance, and venue management. The construction of video big sensor data security application and intelligent algorithm model has become a hot and difficult topic in related fields based on facial expression recognition. This paper focused on the experimental analysis of Cohn–Kanade dataset plus (CK+) dataset with frontal pose and great clarity. Firstly, face alignment and the selection of peak image were utilized to preprocess the expression sequence. Then, the output vector from convolution network 1 and β-VAE were connected proportionally and input to support vector machine (SVM) classifier to complete facial expression recognition. The testing accuracy of the proposed model in CK + dataset can reach 99.615%. The number of expression sequences involved in training was 2417, and the number of expression sequences in testing was 519.

With the rapid development of computer vision and robot technology, smart community robots based on artificial intelligence technology have been widely used in smart cities. Considering the process of feature extraction in fruit classification is very complicated. And manual feature extraction has low reliability and high randomness. Therefore, a method of residual filtering network (RFN) and support vector machine (SVM) for fruit classification is proposed in this paper. The classification of fruits includes two stages. In the first stage, RFN is used to extract features. The network consists of Gabor filter and residual block. In the second stage, SVM is used to classify fruit features extracted by RFN. In addition, a performance estimate for the training process carried out by the K‐fold cross‐validation method. The performance of this method is assessed with the accuracy, recall, F1 score, and precision. The accuracy of this method on the Fruits‐360 dataset is 99.955%. The experimental results and comparative analyses with similar methods testify the efficacy of the proposed method over existing systems on fruit classification.

Industrial Internet of Things is the core field of smart city. And intelligent detection is an important application field of industrial Internet of Things. Demand of the industrial is particularly urgent. In particular, the defect detection of mobile phone shells (MPS) has always been a common problem for famous mobile phone companies. A compression‐free defect detection method (CFDDM) for MPS based on machine vision is proposed in this paper. Firstly, affine transformation is utilized to solve the angle deviation of MPS in different images. Then, edge detection, binarization, and open operation are combined to highlight the edge region based on the results of angle adjustment. It is convenient for region of interest (ROI) extraction and clipping. Finally, the method of gray histogram contrasting is utilized for defect detection according to the results of ROI clipping. And the detection results are obtained. In this paper, MPS data set is utilized for many tests. The results show that the proposed method can effectively detect whether there are defects in MPS data set without image compression. The recognition accuracy is 100%. The recognition time of a single image is about 4.56 s, which is better than other defect detection methods.

The visible light communication (VLC) can provide data transmission via the illumination light emitting diodes (LED). This paper introduces a new model to analyze the bit error rate (BER) of binary phase modulation in VLC for an arbitrary modulation depth and data duty cycle while taking into account both the Gaussian and signal-dependent shot noise. The impact of the driver design on the BER and its impact on ripple have not been considered in detail before. We compare two different LED driver schemes, namely directly adapting the driver control loop and binary shunting. We address data rate, BER, and power efficiency, for which we propose to use the extra energy per symbol above unmodulated light. We further introduce an analysis of the effect that (truncated) ripple interference has in the (matched or other) filter of the receiver. Ripple causes a harmful interference in VLC, and thus a BER expression is derived to include its effect. Two approximations are proposed to model the ripple interference, and their accuracies are compared by simulations. A low-pass filtering is proposed to alleviate the impact of ripple interference in VLC system.

A complete description of trust relationship is key to construct a high precision trust model. But most of existing models not only miss the negative information and the hesitation information of trust, but also ignore the discordance between text comments and ratings. To solve the problems, a dynamic trust evaluation model based on the affect intensity is proposed. In the model, the intensity of sentimental polarities are calculated from words in comments. The corresponding relation between evaluations of trust property and the emotional intensity vector can be also described. Time series weights to distinguish the importance of transaction at different time are determined by the inverse form of exponential distribution. So the dynamic attenuation of trust can be described. To improve the polymerization ability of trust information, the local trust, the feedback trust and the overall trust are calculated by operators of fuzzy logic. The experimental results show the proposed model can effectively describe the trust relationship between nodes to identify and eliminate various malicious attacks significantly. Copyright © 2016 John Wiley & Sons, Ltd.